AI Prior Art and Obviousness at the PTAB: Key Issues for Patent Owners and Challengers

Introduction : Artificial intelligence is no longer merely the subject matter of patents; it is increasingly shaping how prior art is found, assessed, and used to challenge patents. At the US Patent Trial and Appeal Board (PTAB), AI-related inventions are being tested through familiar legal standards - novelty, obviousness, claim construction, and enablement - but the technology’s speed, opacity, and cross-disciplinary character create new evidentiary and strategic challenges. For patent owners and challengers alike, success at the PTAB will depend less on labelling a dispute as an “AI case” and more on rigorously proving what a person of ordinary skill in the art would have known, understood, and been motivated to do.

Why AI Prior Art Is Different

AI patents commonly cover machine-learning architectures, training methods, data-processing pipelines, model optimisation techniques, computer-vision systems, natural-language processing, and AI-enabled improvements to hardware or software performance. The relevant prior art may therefore include academic papers, open-source repositories, technical documentation, conference presentations, product manuals, and earlier patent filings.

At the PTAB, inter partes review (IPR) and post-grant review (PGR) proceedings are limited to prior art consisting of patents or printed publications. This makes the public accessibility, date, content, and enablement of each reference critically important. A reference cannot simply be described as “AI prior art”; the challenger must establish that it was publicly accessible before the relevant critical date and that it actually discloses the claimed limitation. 

AI also complicates the identification of the person having ordinary skill in the art. An AI invention may draw on computer science, statistics, electrical engineering, linguistics, biology, or domain-specific industry knowledge. The proper level of skill can materially affect both the interpretation of references and the obviousness analysis.

The PTAB’s Obviousness Framework

Obviousness under 35 U.S.C. Section 103 asks whether the claimed invention, viewed as a whole, would have been obvious to a person having ordinary skill in the art at the time of the invention. The analysis typically considers:

• The scope and content of the prior art.

• The level of ordinary skill in the art.

• Differences between the claimed invention and the prior art.

• Objective indicia of non-obviousness, such as commercial success, long-felt need, failure of others, and unexpected results.

The PTAB applies this framework through claim construction and detailed mapping of claim limitations to prior-art disclosures. In AI cases, this often turns on whether the reference discloses not merely a general use of machine learning, but the specific model, training approach, architecture, input-output relationship, or technical improvement claimed in the patent.

A useful recent example is Tesla, Inc. v. Autonomous Devices, LLC, IPR2023-01173, where the PTAB issued a final written decision on 3 January 2025 invalidating all challenged claims of U.S. Patent No. 11,055,583. The decision illustrates that AI patents are not insulated from invalidity merely because they involve sophisticated technology; the decisive question remains whether the prior art and permissible motivations to combine render the claimed invention obvious.

AI as a Source of Prior Art

AI-generated material raises difficult questions about whether it qualifies as a “printed publication.” A document generated by an AI system may be prior art only if it was publicly accessible before the critical date and enabled a person of ordinary skill to make and use the disclosed invention. Public accessibility is not established merely because data existed on a server, in a training corpus, or within a private model.

The Federal Circuit has repeatedly emphasised that the party relying on a reference bears the burden of proving public accessibility when that status is reasonably challenged. In the AI context, challengers should therefore preserve evidence of indexing, searchability, public dissemination, download records, repository timestamps, and the identity of persons in the relevant field who could reasonably locate the disclosure. 

For patent owners, the practical message is clear: challenge the evidentiary foundation of AI-generated or AI-discovered prior art. Ask whether the reference was truly publicly available, whether its date is proven, whether it enables the claimed invention, and whether it discloses each claim limitation with reasonable specificity.

Does AI Raise the Level of Ordinary Skill?

One of the most consequential unresolved issues is whether widespread access to AI tools should raise the presumed skill level of the PHOSITA. If AI tools make routine tasks easier - such as literature searching, code generation, data analysis, or model tuning - challengers may argue that inventions once considered inventive are now obvious.

The USPTO recognised the significance of this issue in its April 2024 Request for Comments, which sought public input on AI’s impact on prior art, the knowledge of a PHOSITA, and patentability determinations. The debate remains active: some stakeholders argue that AI should increase the skill level and thereby make more inventions obvious, while others contend that the legal inquiry should remain anchored to the knowledge and capabilities actually available to skilled artisans at the relevant time. 

For now, patent owners should resist sweeping assertions that “AI makes everything obvious.” The better approach is to define the PHOSITA by the specific technology, identify the knowledge reasonably attributable to that person, and show that the claimed technical solution was not a predictable application of known tools.

Motivation to Combine AI References

Many AI patents are challenged through combinations of references: one disclosing a machine-learning model, another disclosing a data-processing method, and a third disclosing an application in a particular field. The PTAB must still find a rational underpinning for combining the references.

A challenger should do more than assert that both references relate to AI. It should explain:

• Why a skilled artisan would have selected the specific references.

• What problem each reference addresses.

• How the references can be combined without changing their fundamental principles.

• Why the combination would yield the claimed result with a reasonable expectation of success.

• Why the claimed improvement, if any, would not have discouraged the combination.

This is particularly important where the patent claims a technical improvement - for example, reduced computational load, improved model accuracy, lower latency, better memory usage, or more efficient training. A patent owner can argue that the prior art taught away from the claimed solution, that the combination would have required impermissible retrofitting, or that the claimed result was unpredictable.

Technical Improvement Still Matters

AI and software inventions face recurring eligibility scrutiny, but recent USPTO Appeals Review Panel practice shows that claims describing concrete improvements to the functioning of a computer or machine-learning system may avoid a Section 101 rejection. In one appeal, the panel vacated the PTAB’s abstract-idea rejection, finding that the claims described a concrete improvement to how machine-learning systems operate rather than merely “math on a computer.” The claims nevertheless remained rejected as obvious.

That outcome carries an important strategic lesson: patent eligibility and obviousness are distinct. Even where an AI invention is framed as a technical improvement, the applicant or patent owner must still demonstrate that the specific advance was not already disclosed or rendered obvious by the prior art. 

Practical Strategy for Patent Owners

Patent owners should treat AI patents as high-risk assets in post-grant proceedings and prepare defensively from the drafting stage.

• Draft claims around concrete technical improvements, not broad outcomes such as “using AI to improve accuracy.”

• Include specific architectures, training procedures, loss functions, data structures, preprocessing steps, or hardware interactions where they support patentability.

• Record evidence of unexpected results, performance benchmarks, and failures of conventional approaches.

• Preserve inventor testimony, lab notebooks, development records, and contemporaneous evidence of the problem being solved.

• In an IPR, scrutinise the public-accessibility evidence for every AI-generated, web-based, or repository-based reference.

• Contest obviousness by showing that the prior art does not teach the claimed technical mechanism or that a skilled artisan would lack a reason to combine the references.

Practical Strategy for Challengers

Challengers should avoid generic “AI was known” arguments. PTAB judges expect a precise claim-by-claim invalidity case.

• Identify references that disclose the claimed model, method, or system with particularity.

• Establish public accessibility through concrete evidence, not inference.

• Define the PHOSITA realistically and support the level of skill with expert evidence.

• Build a coherent motivation-to-combine story tied to the claimed problem and solution.

• Use claim construction strategically to narrow functional or result-oriented limitations.Address objective indicia early, especially where the patent owner relies on commercial success or industry adoption.

Key Takeaway

AI is transforming patent challenges at the PTAB, but it has not replaced the fundamentals of patent law. The central questions remain whether the prior art was publicly accessible, whether it discloses the claimed invention, whether a skilled artisan would have been motivated to combine references, and whether the claimed advance would have been obvious. Patent owners that disclose specific technical improvements and preserve evidence of unpredictability will be better positioned; challengers that build precise, evidence-led obviousness cases will remain formidable.

Author: Vipasha Srivastava in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at  Khurana & Khurana, Advocates and IP Attorney

Endnotes

1. Tesla, Inc. v. Autonomous Devices, LLC, IPR2023-01173 (PTAB Jan. 3, 2025) (final written decision invalidating all challenged claims of U.S. Patent No. 11,055,583).

2. USPTO, Request for Comments on the impact of artificial intelligence on prior art, PHOSITA, and patentability determinations, 89 Fed. Reg. 34217 (Apr. 2024).

3. Lynk Labs, Inc. v. Samsung Electronics Co., holding that a published patent application may qualify as prior art in an IPR as of its filing date.

4. USPTO Appeals Review Panel decision vacating a PTAB Section 101 rejection of machine-learning claims while leaving the obviousness rejection intact.

5. Federal Circuit authority on the burden to establish public accessibility of a purported printed publication.

6. 35 U.S.C. Section 311(b), limiting IPR challenges to patents and printed publications.